Two numbers from the RWS survey this year sit right next to each other, and together they tell you where to stand. Forty-three percent of freelancers and LSPs reported fewer client requests. Meanwhile 74 percent of corporate teams and 76 percent of public-sector teams said their internal demand held steady or grew.
Read those together and the panic story falls apart. The work didn't leave. It walked upstairs into a building you don't have a keycard for.
Where the requests actually went
For years the flow was simple. A company needs Turkmen, they call an agency, the agency calls me. Three links in the chain, and I'm the last one. Now the first link is building its own machine. Enterprises are standing up in-house localization desks, wiring MT into their content pipelines, and handling the easy 70 percent before anyone thinks to email a vendor.
Sixty percent of everybody is running MT now. Eighty percent among LSPs. That's not a forecast, that's the current floor. And 90 to 98 percent of the people using MT are post-editing the output rather than sending it raw. So the volume that used to arrive as "translate this document" now arrives — if it arrives at all — as "the model already did it, clean it up."
That's the squeeze the 43 percent are feeling. Not zero work. Reorganized work. The commodity layer got absorbed by the client's own tools, and what trickles down to the freelancer is either the leftovers or the disasters.
Here's the part nobody says out loud: for a language like Turkmen, the in-house machine is worse than useless on the hard bits. The enterprise desk can run their Spanish and German through DeepL and be 85 percent fine. Run Turkmen through the same pipeline and you get fluent nonsense — grammatically smooth, factually wrong, and confident about it. The desk can't even tell. Nobody on the team reads Turkmen well enough to catch it.
Stop competing for the shrinking pool
The instinct when requests dry up is to chase harder. Answer faster, quote lower, take the MTPE job at half the per-word rate because half of something beats none of nothing. I understand the math. I also think it's a trap you build for yourself.
If you position as the person who cleans up whatever the machine spat out, you've priced yourself as a function of the machine. Your rate is now pinned to how good the MT got last quarter, and it only gets better. That's the post-editing trap the trade blogs keep circling: working harder for roughly half the money per word, on the theory that the machine did the "real" work. It didn't. It did the confident-sounding work.
The better move is to position for the thing the in-house teams can't fake. Eighty-four percent of language service integrators had a client come back in the past year and ask for a human to review and improve AI output. Accuracy worries sat at 72 percent, quality worries at 68 percent. That's not a market that trusts its own machine. That's a market that got burned and now needs someone who can look at a paragraph and say "this is wrong, and here's why, and here's the term that should be there."
That someone charges by judgment, not by the word. Different pricing conversation entirely.
What the new address means for how I pitch
The traditional language services piece of the market contracted 2 percent. The language-technology side grew 12 percent. Widen the lens to AI-adjacent stuff — data creation, annotation, speech and video workflows — and you get 18 percent growth. The money is flowing toward the tech and toward the people who feed and check it. Not toward standard word-count translation.
So when I talk to a PM now, I'm not selling myself as a pair of hands for whatever volume falls off the truck. I'm selling the things an in-house desk structurally can't do for a small language:
- Terminology ownership. Somebody has to decide, once, whether it's the Russian loanword or the Turkmen coinage, and lock it. The enterprise machine won't. It'll flip-flop across the document.
- The reliability read. Tell me what the model already touched and I'll tell you where it lied. For Turkmen that's a service, because the failure mode is invisible to everyone who doesn't read the language.
- The business argument the client already believes. Eighty-seven percent of B2C shoppers won't buy from an English-only site. Sixty-five percent prefer their own language even when they read English fine. The client knows native-quality moves money. My job is to be the credible reason they don't gamble that on a machine.
None of this is about being anti-MT. I run it, I edit it, I take the money. It's about refusing to let the machine set my address. The in-house teams took the easy volume. Fine. They also inherited every hard problem that comes with a small language and no one to check the output. That problem has my name on it, and it doesn't price by the word.
The demand is stable or growing. It just stopped calling the number I'm used to. Time to change the number I answer to.